Wan2.5-Preview Image

Here's the situation this node exists for: Wan is the video family the local community practically lives on - but after Wan 2.2, Alibaba stopped releasing weights. Wan 2.5, 2.6, and 2.7 are all API-only. So the "Preview Image" tier of Wan 2.5 is a model that literally cannot run on your GPU, and this node is the only ComfyUI-shaped way to touch it. No checkpoint hunt, no VRAM math - the model id in the request is a Runware catalog entry, and the images come back as a native IMAGE.

For Wan 2.5 specifically, "preview" is doing real work in the name: it's the early, get-a-taste tier of the 2.5 series rather than the flagship. If you're after the full 2.7 experience, the Wan2.7 Image node is a couple of rows down in the same menu. Use this one when you want to see where Wan's image generation went after the open-weights freeze.

What you set

positivePrompt is required. The surface is deliberately bare:

• width / height - 1 to 4096px, defaults 1024². Keep it sensible; 4096 on a preview-tier model is a fast way to find its ceiling.
• numberResults - up to 20 per run; each one is a billed generation.
• watermark - optional RUNWARE_WATERMARK builder socket.
• advanced_json - the escape hatch, and the one input worth mentioning beyond the obvious: it merges arbitrary JSON into the request under providerSettings. That's how you reach Wan-specific provider options that don't have dedicated widgets yet. If a Runware doc mentions a Wan 2.5 setting you can't find as a widget, it goes here.

Output is image (IMAGE) → Preview/Save Image. outputFormat JPG/PNG/WEBP, outputQuality, and the standard safety/ttl toggles.

Installing

Part of the Runware/ComfyUI-Runware pack.

ComfyUI Manager: search Runware, install, restart.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Deps are runware-sdk, pillow, soundfile - no model files, obviously, since there are no public weights. Runware API key from runware.ai/api-keys (Settings panel or RUNWARE_API_KEY).

Where people get burned

The expectations trap: "Wan" carries a lot of community goodwill earned by the open 2.1/2.2 video models, but this is a closed, preview-tier image model you're renting. Don't assume it inherits the uncensored, Apache-2.0 character of the open family - the API tier has its own moderation and pricing. Treat the first run as reconnaissance: if the "preview" quality isn't obviously above your local Z-Image or Qwen setup, the cheaper local option wins for daily work.